THE USE OF PREDICTIVE MODELLING TO BOOST DEBT COLLECTION EFFICIENCY

Transcription

1 CREDIT SCORING AND CREDIT CONTROL XIII EDINBURGH AUGUST 2013 THE USE OF PREDICTIVE MODELLING TO BOOST DEBT COLLECTION EFFICIENCY MARCIN NADOLNY SAS INSTITUTE POLAND

2 Many executives fear that the reputation of their companies will suffer if collections become too aggressive, but the careful segmentation of debtors and the flexible pursuit of delinquents applying pressure, in other words, only where it s really justified can largely eliminate these risks. Tobias Baer, Rami K. Karkian and Piotr Romanowski The McKinsey Quarterly: Best Practices for Bad Loans November 2007

3 AGENDA Overview Segmentation Predictive modelling Model monitoring issues Collection communication optimization Effective implementation of collection strategies New trends in predictive debt collection

4 OVERVIEW

5 OVERVIEW SAS APPROACH Know which customers are most likely to respond Proactively manage the customer experience Responding rapidly to changing profile of customers with delinquency issues Determine which channels to use for individual customers Optimise your collections strategies Minimizing losses within context of capacity constraints

6 OVERVIEW STEPS TO PREDICTIVE COLLECTIONS AND RECOVERY Data Prep Segmentation / Predictive models Strategy Assignment Model Monitoring / Validation Metrics / Reporting Account-level history Segmentation Scores to strategies Model monitoring Delinquency Cross-product behavioral data Target variables External data Full customer view Data quality New data types and sources Model development Model deployment Scoring generation New analytical techniques Rules-based strategy assignment Communication optimization Champion-Challenger Strategy deployment Strategy execution Model validation Collector efficiency Channel efficiency Charge-Off Trend Strategy effectiveness

7 OVERVIEW SAS INTELLIGENT DEBT MANAGEMENT Management Information Analytics Optimisation Automation Core systems KPI s / Reports Slice & Dice Segmentation Predictive models Propensity to pay Channel selection Campaign Mangt. Campaign Deployment Collections system Variance / Alerts Segmentation Sequence Analysis Text Mining Ops.Constraints Business Objectives Analytics Processes Analytics Deployment Recovery system Future forecast Performance Mgmt SNA Forecast / Trends Best case scenarios Stress Testing Champion / Challenger Model Management Letters, etc Single Customer View C&R Integrate Profile Standardise De-duplicate Augment Allocate costs Core systems Collections system Recovery system Credit Bureau DCA performance Complaints system Other 3rd party

8 SEGMENTATION

9 SEGMENTATION SEGMENTATION CRITERIA 1DPD Segmentation criteria business Consumer SME Corporate operational product behavioural Segment A Segment B Segment C Segment D statistical Early collection 1-30 DPD) Late collection 30+ DPD

10 PREDICTIVE MODELLING

11 PREDICTIVE MODELLING Predictive Modelling Flow Defining Target Variables Analytical Data Mart Model Form Key Predictors

12 PREDICTIVE MODELLING PREDICTIVE MODELLING FLOW Predict future events based on the past Initial analysis Data preparation Univariate analysis Multivariate analysis Validation Model usage & monitoring

13 PREDICTIVE MODELLING DEFINING TARGET VARIABLES PROBABILITY OF CURE Outcome period Descriptive variables Target variable Model In/Out criteria (segmentation) Cure criteria Process and business dependent Amount and time tolerance thresholds Various prediction horizons Aggregation on the client level

14 PREDICTIVE MODELLING LARGE SME & CORPORATES ANALYTICAL DATA MART Financial ratios Behavioural variables ID Qualitative factors Target variables

15 PREDICTIVE MODELLING LARGE SME & CORPORATES ANALYTICAL DATA MART Financial ratios Liquidity Debt Profitability Leverage ratios Efficiency Bankruptcy Target variables ID Macroeconomic situation Market outlook Management quality Organization assessment Cost structure Product portfolio Quality of contrahents Qualitative factors

16 PREDICTIVE MODELLING LARGE SME & CORPORATES ANALYTICAL DATA MART Behavioral variables Products Balances Account transactions Card transactions Overdue characteristics Target variables ID Collaterals Contact history On-line behaviour Demography External data

17 PREDICTIVE MODELLING CALCUTATION OF BEHAVIOURAL CHARACTERISTICS Monthly data Monthly data Mapping formulas F[x] Analytical data Average Weighted Average Trend Trend upward and downward Minimum Maximum Sum Count Unique Count Ranks Logic Sum

18 PREDICTIVE MODELLING LARGE SME & CORPORATES MODEL FORM Financial ratios Financial model Qualitative factors Qualitative model Combined model Collection strategy Behavioural data Behavioural model

19 Financial Qualitative Behavioural PREDICTIVE MODELLING LARGE SME & CORPORATES KEY PREDICTORS Assets Debt ratios Profitability ratios Efficiency ratios Liquidity ratios Management quality Enterprise history History of cooperation with banks Months overdue Number of credits overdue Principal / Interest overdue amount

20 MODEL MONITORING ISSUES

21 MODEL MONITORING MODEL MONITORING ISSUES What? Predictive power Stability Calibration quality (at strategy level) Outcome period Predictive power Calibration quality Stability

22 MODEL MONITORING ISSUES Issue - Low number of cases results in calculation difficulties Solution cumulate cases from several months Outcome period (i) i i+1 i+2 i+3 i+4 i+5 Cumulate cases entering the collection process

23 COLLECTION COMMUNICATION OPTIMIZATION

24 PREDICTIVE DEBT COLLECTION COLLECTION OPTIMIZATION PROCESS Define Collections Strategies Mathematical Optimisation, based on linear programming offers superior performance over common rule-based prioritisation methods Customer Data Define Optimization Scenarios Optimize Examine Optimization Reports Execute Optimal Collections Predictive scores by channels

25 EFFECTIVE IMPLEMENTATION OF COLLECTION STRATEGIES

26 IMPLEMENTATION OF STRATEGIES COLLECTION STRATEGIES Delinquent customers High Propensity to Collect Medium Propensity to Collect Low Propensity to Collect Previous Defaulter Low Debt High Debt Champion Challenger Soft Letter Medium Call & Review Medium Letter & Review Hard Call and Review Refused OK OK Refused Refused OK OK Refused Medium Call Trace Hard Call Trace Field Visit

27 IMPLEMENTATION OF STRATEGIES STRATEGY DEFINITION AND EXECUTION Strategy definition Strategy rules Predictive model deployment Champion/Challenger Multi channel Multi step paths Strategy simulation/execution Data for treatment paths Strategy simulation, what-if analysis Channel execution Contact history creation Next stage definition Rapid deployment of strategies

28 NEW TRENDS IN PREDICTIVE DEBT COLLECTION

29 PREDICTIVE DEBT COLLECTION NEW TRENDS SOCIAL NETWORK PREDICTORS TEXT MINING CALL CENTER NOTES BIG DATA REAL-TIME

30 BENEFITS SELECTED BENEFITS FROM PREDICTIVE DEBT COLLECTION Lower cost of debt collection Higher recoveries Less court litigation Higher contactability Shorter time to recovery Significantly larger volume in collections Improved collectors productivity Higher customer satisfaction

31 THANKS